You are on page 1of 4

GEOMETRICAL PARADOXES

(50-MINUTE TALK FOR MATHCAMP)


MD
1. Intro: The Infinite Hotel
Statement of standard Banach-Tarski paradox, as a preview:
Let B = B
1
(0) R
3
be given by B = x R
3
: |x| = 1. Then there are sets
A
i
and rigid motions g
i
so that B = A
1
.A
2
.A
3
.A
4
.A
5
and

g
i
A
i
= B

.B

for B

= B
1
(x

), B

= B
1
(x

).
In fact, if P = B
teeny
(0) (to be thought of as a pea) and S = B
huge
(0) (the
sun), there is a partition of P into nitely many pieces (where the number depends
on the ratio of huge to teeny) and there are rigid motions so that the rearranged
pieces of the pea make up the sun.
Recall: sets, countability
B-T will turn out to rely critically on a version of the Innite Hotel paradox.
(An N-numbered hotel can accommodate: one more; 35 more; N more; N
2
more.
But the rst two can be done rigidly and the last two can not.)
2. A Paradoxical Null Set in the Plane
Recall: a metric space is (X, d) for a distance/metric d : X
2
R
0
. An isome-
try f : X X

is a map of metric spaces that preserves distance: d


X
(f(x), f(y)) =
d
X
(x, y). Composition of functions gives a group law (identity, inverses and the rest
check out) so the full collection of isometries Isom(X) is called the isometry group.
Historical Note: Felix Klein in the 1880s posited the (then radical) idea that a
denition of geometry starts with a transformation group and considers its invari-
ants: that is, geometry is what stays the same under change, say by symmetries.
This puts the isometry group at least on equal footing with the pair (X, d) as a
way of understanding a geometry.
Denition of (nite) equidecomposability: for A, B X, the notation A B
means that there exist partitions of A, B, and isometries g
i
Isom(X), such that
A = A
1
. . A
n
, B = B
1
. . B
n
, and B
i
= g
i
A
i
for all i.
Call a set B paradoxical if B 2B. This will be equivalent to the following, which
is easier to check: having two subsets C, D such that C . D B and C B D.
With this denition, we can see that none of the Innite Hotel maneuvers is a
geometrical paradox on its own: the rst two are accomplished by rigid motions,
but they do not double N, while the third does give a way of doubling N but it
requires a non-distance-preserving map. But this is not just bad luck; it turns out
to be impossible to concoct a geometric paradox on N with the metric it inherits
from the real line.
Fact: There is no nonempty subset of R which is paradoxical.
1
2 MD
You could think about this in terms of the metric space as not having enough
room to arrange the points in a suitable way, or (to be Kleinian about it) as
the isometry group of R (generated by just one translation and one ip) being too
small. Moving up from the line to the plane gives us more room and many more
isometries.
Construction: an innite paradoxical set in R
2
(the Sierpinski-Mazurkiewicz
Paradox):
Choose so that it is transcendentally related to 2 (eg, = 1 works, as do
most values). Then let H = a
0
+a
1
e
i
+a
2
e
2i
+ +a
n
e
ni
[a
i
, n Z
0
.
If H
0
= h H : a
0
= 0 and H
+
= h H : a
0
> 0, then H = H
0
. H
+
,
H =
1
H
0
, and H = t
1
H
+
, where is multiplication by e
i
and t is addition of
1.
Note: this paradoxical set has measure zero (it is a null set). It turns out
that its not possible to come up with a paradoxical decomposition of any set with
interior in the plane. To do that, you need to go up to R
3
, whose isometry group
is again much bigger.
3. The Free Group, its Tree, and its Paradox
Recall: trees, groups, generators, words
The (non-abelian) free group is F
n
= a
1
, . . . , a
n
[). (Note that if you abelianize
you just get F
ab
n
= Z
n
.) Here, F
2
will play a very central role.
F
2
is represented by 4-regular tree (which well call T
4
): label a point e for the
identity element, and let each application of a, b, a
1
, b
1
correspond to , , , ,
respectively. We can make this tree a metric space by giving every edge length one.
T
4
can be drawn symmetrically in the plane (as a kind of fractalline structure) by
drawing the edges farther from e smaller and smaller, but remember that the real
tree is homogeneous (the same at every point).
Notice that F
2
acts on itself isometrically by left-multiplication: for instance, a
3
moves everything three positions to the right, which is a distance-preserving map.
We can decompose the tree by
F
2
= e . W
a
. W
a
1 . W
b
. W
b
1,
where W
a
is the collection of words whose rst letter is a (so it is the right side of
the tree) and likewise W
b
, W
a
1, W
b
1. (The rst letter of a word in the free group
is well dened because there are no relations, so any two words that look dierent
are truly inequivalent.)
But F
2
= W
a
. aW
a
1 and F
2
= W
b
. bW
b
1, so the free group is paradoxical
by its self-action.
So looking back at null-set paradox in R
2
, we found two rigid motions t and
which generated a free subgroup in the isometry group (that is, a copy of F
2
). It
turns out that free actions are the great fountain of paradoxes.
4. The Banach-Tarski Paradox
Recall: actions, orbits, the axiom of choice (AC). Below, a xed point of a group
will be a point xed by any group elementnot necessarily by all.
Fact: Whenever F
2
acts on a space without xing any points, the space is
paradoxical with respect to that action.
GEOMETRICAL PARADOXES (50-MINUTE TALK FOR MATHCAMP) 3
(Idea: then the orbits are copies of the tree T
4
and they decompose as before.
The lack of xed points means that the trees retain their structureno two points
are identied. Since every point is in some orbit, the collection of trees partitions
the whole metric space. In general, putting these together into sets in the equide-
composition requires the choice of one representative point from each orbit to play
the role of the center of the tree: this is where AC is used.)
Fact: Most pairs of rotations of R
3
generate a free group. (That is, such pairs
are dense in SO(3) SO(3).)
This equips us to build a paradoxical decomposition of the sphere, from which
well put together a paradox on the ball. Each rotation has two xed points on
the sphere (the intersection of the sphere with the axis of rotation). But then the
group generated by the two independent rotations has only countably many xed
points. So it follows from the Fact that S
2
D is paradoxical for some countable
set D.
Proposition: For any countable subset D of the sphere S
2
, there is a two-piece
equidecomposition by rotations S
2
S
2
D.
Proof: Find a rotation of the sphere so that all of the sets
i
(D) are disjoint.
(Possible because D countable, so at each stage only countably many rotations will
hit the previous sets.) Then let =

i=0

i
(D), which is also countable. Clearly
S
2
= (S
2
) . . Now note that () = D. This is just the Innite Hotel at
work! Applying moves everybody over one room, opening up the rst.
= D. (D) .
2
(D) .
() = (D) .
2
(D) .
Therefore
S
2
= (S
2
) . (S
2
) . () = S
2
D.

Finally, we extend this decomposition radially to give a paradox on B 0. We


have seen that a whole countable set can be absorbed by equidecomposition, so
the missing origin is little obstacle!
5. Cool Closing Points
B-T has a stronger form: any two bounded subsets of R
3
with interior
are equidecomposable! (Hence the pea-into-sun rearrangement.) This is
achieved by covering with balls.
What does this mean about volume? In a way, these are fractal pieces,
which have no volume dened for them. What does this mean about mea-
sures? The nonexistence of certain really good measures: the sets in the
partition must be unmeasurable with respect to Lebesgue measure, and in
fact there must not exist any measure which is dened on all subsets of R
n
.
Free groups produce paradoxes, but paradoxes dont require free groups:
Grigorchuk and Olshanskii have an example of a paradoxical group with
only nite-order elements, which clearly has no free subgroup.
What does this mean about groups? Paradox-producing-ness turns out to
be related to growth rate: F
2
grows really fast (exponentially) and this
helps. There are strong relationships between growth rate and other group
theoretic properties.
4 MD
Contrary to many rumors, the Axiom of Choice is not essential to geometric
paradoxes: on one hand, we didnt need it for the null-set paradox in the
plane. Also, the hyperbolic plane, for instance, has H = A . B . C with
A B C and A B C (so that the subset A is at once a half and a
third of H); no choice needed.

You might also like